Optimization of native buffer accesses in JAVA applications on hybrid systems

Abstract

Managing buffers in a hybrid system, in one aspect, may comprise selecting a first buffer management method from a plurality of buffer management methods; capturing statistics associated with access to the buffer in the hybrid system running under the initial buffer management method; analyzing the captured statistics; identifying a second buffer management method based on the analyzed captured statistics; determining whether the second buffer management method is more optimal than the first buffer management method; in response to determining that the second buffer management method is more optimal than the first buffer management method, invoking the second buffer management method; and repeating the capturing, the analyzing, the identifying and the determining.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS[0001]This application is a continuation of U.S. Ser. No. 13 / 732,820, filed on Jan. 2, 2013, the entire content and disclosure of which is incorporated herein by reference.FIELD[0002]The present application relates generally to computers, and computer applications, and more particularly to managing buffers in a hybrid system.BACKGROUND[0003]Java Virtual Machine (JVM) implementations support the Java Native Interface (JNI) as a mechanism which allows Java code to call methods written in C and C++(native code) and vice verse. Traditionally both the code written in Java and the native code is executed in the same process and by the same thread as execution transitions between the two.[0004]It is possible, however, to construct a JVM such that the native code is run in one or more Remote Execution Containers which may be hosted in separate processes on the same or different machines from where the Java code is executed such that the native code is ...

AI Technical Summary

Benefits of technology

This patent describes a method for managing a buffer in a hybrid system that includes both native code and managed code. The method involves selecting a buffer management method from a list of options, capturing statistics about the buffer while running under the selected method, analyzing the statistics to identify a more optimal buffer management method, and invoking the more optimal method if it is available. The technical effect of this invention is an improved performance and efficiency of buffer management in hybrid systems.

Problems solved by technology

In a split JVM, the cost of the calls between Java and native code has much greater overhead and latency resulting in the need to reduce round-trips where possible.

In the case of a distributed JVM, however, the latency of making a cross-process / cross-machine call may be magnitudes of order greater than required to run the method called.

When a native buffer is accessed from both Java and native methods, accessing the buffer from remote machine will cause a network round-trip overhead.

This naïve allocation of direct byte buffers on the JNI-side (host-side) makes accesses expensive for some cases.

Excessive roundtrips for data make this approach non-usable for real applications in which slowdowns have been observed in excesses of 50 times.

Images